Information retrieval system based on Romanian continuous speech recognition

This paper presents practical results concerning the implementation of an experimental information retrieval system based on continuous speech recognition in Romanian. The main task of the platform is to allow students to access and to retrieve information from a database via an automatic speech recognition system that uses continuous hidden Markov models. The application´s architecture is composed by a MySQL database with students´ information and an engine that transforms the speech commands uttered in continuous manner into a query string for the database. The machine´s feedback is not yet a text to speech one. For experimental purposes the dialogue is based on a simple language model in order to get the appropriate sequence of commands. This system was our first trial in the application of continuous speech recognition for a practical purpose aimed to identify inherent problems and to see how language models could be incorporated.